It seems in the article the author suggests that in certain case tries is faster than hash table.

Yes, you are right about every data structure can be badly implemented. But some data structures are more easily to get wrong. For example, everyone have no problem with selection sort, it is just hard to get it wrong. But for binary search, some decent programmer may write (a + b) / 2. Same case for tries, people more commonly (intuitively) write things like node_t* child[26];